Average-Pooling Temporal Graph Convolutional Network for Anomaly Detection of Electric Submersible Pump

The electric submersible pump (ESP) plays a vital role in oil production, and precise detection of ESP anomalies can significantly improve both safety and operational efficiency. This paper proposes a novel approach for ESP anomaly detection, referred to as the average-pooling temporal graph convolutional network (APTGCN). APTGCN integrates a temporal convolutional network (TCN) and graph convolutional network (GCN) to model spatio-temporal dependencies among ESP monitoring variables. Particularly, an average pooling layer (APL) is employed between the TCN and GCN layers, which serves to reduce computational complexity and enhance the efficiency of training. Extensive experiments on real ESP data collected from the oil field indicate that APTGCN outperforms other state-of-the-art methodologies for ESP anomaly detection.